feat(migrate): add Graviton (ARM64) cost optimization#86
Open
herosjourney wants to merge 17 commits into
Open
Conversation
Expand generate-artifacts-report.md with combined TCO, security baseline component costs, gap analysis, and assumptions sections; add an HTML validator script, reference fixture, and tests so stub appendices fail validation instead of shipping as complete reports. Co-authored-by: Cursor <cursoragent@cursor.com>
…ults. Fix broken TOC anchors in the reference fixture; validate nav links match section ids; require exactly one <section id> per required block; scope GuardDuty checks to cost/security appendices; gate exec-tco on both estimate JSON files; default to rename-on-fail; expand test coverage to 11 cases. Co-authored-by: Cursor <cursoragent@cursor.com>
…RT_OK scope. Validate reference HTML with estimation-infra/ai fixture JSON; replace dead GuardDuty regex with component dollar matching; document that REPORT_OK means structure complete not numerically audited; expand tests to 14 cases. Co-authored-by: Cursor <cursoragent@cursor.com>
Add Generate-phase report overview, REPORT_OK scope, pytest/validator commands, and fix stale skill/docs paths in Development section. Co-authored-by: Cursor <cursoragent@cursor.com>
Replace the hardcoded SF Beach path with STUB_FAIL pytest coverage, add a committed stub fixture for manual CLI validation, and use word-boundary regex in _dollar_amount_present so $13 does not match $130. Co-authored-by: Cursor <cursoragent@cursor.com>
Drop project-specific structural-reference wording from the PR-facing README and reference HTML footer. Co-authored-by: Cursor <cursoragent@cursor.com>
Add readability gates (no Rubric:/Section N headings), rewrite the reference report with security teaser + appendix split, and document conventions in the spec so generated reports stay exec-friendly. Co-authored-by: Cursor <cursoragent@cursor.com>
Require exec-security-teaser when security_baseline exists, a verdict banner when recommendation exists, and --migration-dir canary checks on real runs; refresh the stub regression fixture and wire --migration-dir into Generate Step 4. Co-authored-by: Cursor <cursoragent@cursor.com>
Reframe Stay if, risk severity, deferred callout, and estimate/report spec so phased infra+AI migration proceeds while analytics target is evaluated in parallel with AWS account team or data partner. Co-authored-by: Cursor <cursoragent@cursor.com>
Correct observability messaging (5 GB logs / 10 metrics / 10 alarms always-free vs GCP's larger tier), subtract allowances in Step 4, and recascade SF Beach reference totals ($118→$112 infra, ~$503 combined savings). Co-authored-by: Cursor <cursoragent@cursor.com>
Ensure user-facing cost output consistently uses 'estimated monthly costs' phrasing across the estimate phase summaries, report template, README, and report fixture. Adds an explicit cost-labeling rule to the estimate orchestrator and each sub-estimate Present Summary.
…xec flow Apply Anthropic frontend-design skill principles to the migration report: - "Name things by what people control, not how the system is built": add a post-write gate so executive-flow sections (decision-summary, exec-*) cannot expose artifact filenames (*.json) or Terraform resource IDs (aws_<resource>.<name>). Those identifiers stay in the technical appendices, which remain exempt. Gated under --no-readability. - "Structure is information": clarify the numbered-heading ban targets decorative labels only — genuine sequences (cluster order, phased weeks, migration phases, rollback steps) keep their numbering. - "One name per concept": require a single consistent label for the recommended model and cost tier across verdict, tables, and appendices. Updates the reference fixture to reader-facing copy in exec-tco and exec-security-teaser, documents check awslabs#14 + the sequence nuance in validate-migration-report.md and generate-artifacts-report.md, and adds 4 tests (28 pass).
Adds Graviton/ARM64 as the default architecture for eligible compute, wired through all six phases of the gcp-to-aws skill. New canonical references: - shared/graviton.md — compatibility tiers, GCP→Graviton mapping, per-phase rules - shared/schema-graviton.md — graviton_profile, cpu_architecture, architecture_comparison Phase wiring: - Discover (app-code + IaC): emit graviton_profile per compute service - Clarify: Q11b — conditional Graviton question; auto-default when all services are tier=ready - Design (compute): branch on design_constraints.cpu_architecture; emit graviton block - Estimate: Part 2C architecture_comparison (hourly price discount only) + EC2 dual-arch MCP recipe - Generate: emit runtime_platform ARM64 / architectures=["arm64"] / Graviton instance types Pricing cache: add verified us-east-1 ARM64 rows (t4g/m7g/c7g/r7g, Fargate ARM64, Lambda arm64); uncached sizes fall back to the awspricing MCP. SKILL.md: register graviton.md as a conditional reference (compute/db/cache present or graviton_profile exists); add Graviton CPU-architecture default. Scope: gcp-to-aws only. Heroku parity and the report numeric cross-check (pairs with PR awslabs#78) are follow-ups.
…fy Graviton logic (a) Report/validator honesty: - Soften the over-stated claim that the report validator enforces Graviton numeric consistency. PR awslabs#78's validate-migration-report.py is a structural/readability gate and explicitly does NOT audit dollar figures. - Document architecture_comparison as an optional field in schema-estimate-infra.md (additive; outside PR awslabs#78's edited region). - Specify report rendering + a validate-migration-report.py numeric assertion as an explicit follow-up that lands on top of PR awslabs#78 (graviton.md "Report rendering"). No edits to PR awslabs#78-owned report/validator files. (c) Clarify fallbacks: - Replace Q11b fire/skip prose with an exhaustive decis- Replace Q11b fire/skip prose with an exhaustive decis- Replace Q11b fire/skip prose with an exhaustive decis- Replace Q11b fire/skip prose with an exhaustive decis- Replace Q11b fire/skip prose with an exhaustive decis- Replace Q- Define "risk signals" precisely, mirroring schema-graviton.md detection tables. Avoids collision with open PR awslabs#78 (no shared report/validator file edits).
Addresses second Cursor pass: 1. clarify.md orchestrator now registers Q11b — phase index (Q8–Q11b), Category C question list, and a validation-checklist item requiring cpu_architecture when compute is present (written even when auto-defaulted). 2. clarify-compute.md Q11b decision table: explicit row for IaC-only profiles (source=iac, no app_code profile; e.g. machine_type-only conditional) → ask, removing the strict-agent ambiguity. 3. Document database/cache cpu_architecture propagation as a tracked v1 limitation (managed DB/cache stay Graviton-default; no compatibility risk — consistency follow-up). 4. Mark billing-only graviton_profile as planned/not-yet-emitted in schema-graviton.md; Q11b row 1 covers billing-sourced compute meanwhile. 5. compute.md Example 3: t3.medium → t4g.medium (Graviton default). No edits to PR awslabs#78-owned files.
…Notes Closes the load-condition inconsistency surfaced while tracing outputs: - SKILL.md conditional-load condition for graviton.md now reads "Design, Estimate, Generate" (was "Design and Estimate only") — so the Generate phase is authorized to load it, matching the existing reference to graviton.md in generate-artifacts-infra.md. - graviton.md top note updated to match. - generate-artifacts-docs.md: add a "Graviton Migration Notes" subsection (Section 4, infra track) — services moved to arm64, conditional caveats, the arm64 docker build step, and the load-test-before-downsizing note. Renders only when a compute resource targets arm64; skipped on x86 opt-out. Result: the Graviton arm64 guidance now reliably reaches the generated MIGRATION_GUIDE, not just the Terraform.
Optional polish from review: add a Q11b (CPU architecture) row to the Defaults Table with the auto-default values, and name Q11b explicitly in the Batch 2 planning table. The Step 5 structural checklist is left as-is since the handoff Validation Checklist gate (#4) already enforces cpu_architecture when compute is present.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Depends on
Summary
Add Graviton (ARM64) cost optimization to the gcp-to-aws migration skill. When compute workloads are eligible for AWS Graviton processors, the skill now defaults to ARM64 instance types (~15–20% hourly price savings) and surfaces the decision appropriately through the migration phases.
Problem
The migration skill always emitted x86 instance types regardless of workload compatibility. Graviton instances offer 15–20% lower per-hour pricing at the same vCPU/memory spec, but the skill had no mechanism to detect eligibility, ask the user when ambiguous, or emit ARM64 targets in Terraform output.
Solution
graviton.md(single source of truth for tiers, mappings, phase behavior) andschema-graviton.md(JSON schemas forgraviton_profile,architecture_comparison, design block)graviton_profileper compute service from app code signals (Step 2.7 indiscover-app-code.md) and IaC signals (discover-iac.md)ready; only asks whenconditional/unknownwith risk signalscpu_architecturepreference to emit Graviton or x86 instance families per servicearchitecture_comparisonblock inestimation-infra.json; add Graviton rows topricing-cache.mdruntime_platform, Lambdaarchitectures, EC2 Graviton families); add "Graviton Migration Notes" to output docsgraviton.mdconditional loading ruleCompatibility tiers
Design decisions
ready— consistent with existingdb.t4gdefaultsTest plan
db.t4g, Redis → ElastiCachecache.t4g)conditional/unknownprofiles with risk signals existreadyestimation-infra.jsonemitsarchitecture_comparisonblock with correct savings mathFiles changed (13)
All within
migrate/plugins/migration-to-aws/skills/gcp-to-aws/:SKILL.md— conditional load rulereferences/shared/graviton.md— newreferences/shared/schema-graviton.md— newreferences/shared/pricing-cache.md— Graviton pricing rowsreferences/shared/schema-estimate-infra.md—architecture_comparisonschemareferences/phases/discover/discover-app-code.md— Step 2.7references/phases/discover/discover-iac.md— Graviton signals from machine_typereferences/phases/clarify/clarify-compute.md— Q11b decision tablereferences/phases/clarify/clarify.md— batch index + defaults tablereferences/design-refs/compute.md— branch on cpu_architecturereferences/phases/estimate/estimate-infra.md— architecture_comparison emissionreferences/phases/generate/generate-artifacts-infra.md— ARM64 Terraformreferences/phases/generate/generate-artifacts-docs.md— Graviton Migration Notes section